From the article:

"One major caveat is that Project Mariner only works on a Chrome browser's foremost active tab, which means you can't use your computer for other things while the agent works in the background – you need to watch Gemini slowly click around."

Web / GUI agent implementations will have to be moved off the local device to ever be useful, otherwise they block the user's machine. I imagine eventually apps using web / GUI agents internally may abstract away the "browsing live view" entirely - instead of having users watch an agent work in real-time, the agent would run asynchronously in the cloud and just return the final outcome or report.

I'm working on an API for AI agent virtual desktops, so thinking through this a lot currently! https://www.agentstation.ai/